Optional delay for streamed replies nobody is watching
Setting CLAUDE_CODE_UNWATCHED_REPLY_HOLD_MS makes Claude Code batch streamed replies for subagents, side requests and compaction instead of handling each piece
Under the hoodTier: how much it should matter to you
2Useful: my rating, 1 to 5
3Signal: worth watching, 1 to 5
StreamingArea: what it touches
Internal ChangesKind: in v2.1.311,
Internal ChangesSection of the release
What
Claude's replies arrive as a stream of small pieces. Some requests run in the background where no one is watching the text appear:
subagents, which are helper agents Claude starts to do part of a task
auxiliary requests, which are small side calls Claude Code makes for itself
compaction, where Claude Code summarises a long conversation to save space
For these, Claude Code can now hold the streamed pieces back and pass them along in batches. The hold time, in milliseconds, comes from the environment variable CLAUDE_CODE_UNWATCHED_REPLY_HOLD_MS. It defaults to 0, which means no holding at all, so nothing changes unless you set it.
Why
Batching background streams can mean less work for your computer while nobody is reading them. It stays off unless the variable is set above 0.
